Photoplethysmography and Deep Learning: Enhancing Hypertension Risk Stratification

Area of Science:

• Cardiovascular Physiology
• Biomedical Signal Processing
• Artificial Intelligence in Medicine

Background:

• Blood pressure monitoring is crucial for cardiovascular health.
• Early detection of hypertension is vital for preventing related diseases.
• Current methods for hypertension risk stratification can be improved.

Purpose of the Study:

• To evaluate deep learning's effectiveness in hypertension risk stratification.
• To compare deep learning with classical signal processing methods.
• To utilize photoplethysmography (PPG) signals for hypertension assessment.

Main Methods:

• A deep learning model (GoogLeNet) was trained using PPG signals.
• Continuous wavelet transform (Morse) was applied for feature extraction.
• Data from the MIMIC Database (121 recordings) including arterial blood pressure (ABP) and PPG signals were used.
• Hypertension categories: normotension (NT), prehypertension (PHT), and hypertension (HT).
• Classification trials: NT vs. PHT, NT vs. HT, and (NT + PHT) vs. HT.

Main Results:

• The deep learning method achieved F-scores of 80.52% (NT vs. PHT), 92.55% (NT vs. HT), and 82.95% ((NT + PHT) vs. HT).
• Deep learning demonstrated superior accuracy compared to classical signal processing methods.
• The approach showed comparable results to methods using both electrocardiogram and PPG signals.

Conclusions:

• Deep learning offers a promising, accurate approach for hypertension risk stratification using PPG signals.
• This method enhances early detection and assessment of hypertension.
• The findings suggest a potential non-invasive tool for cardiovascular health monitoring.